WP-2019-020

The Women’s Empowerment in Nutrition Index

Sudha Narayanan, Erin Lentz, Marzia Fontana, Anuradha De, and Bharati Kulkarni

Indira Gandhi Institute of Development Research, MumbaiJune 2019

The Women’s Empowerment in Nutrition Index

Sudha Narayanan, Erin Lentz, Marzia Fontana, Anuradha De, and Bharati

Kulkarni

Email(corresponding author): sudha@igidr.ac.in

AbstractRural women face many obstacles that thwart their well-being. Policies that seek to empower them, for

example, by improving livelihood opportunities, often do not translate into improvements in other areas,

notably in their nutritional status. Indeed, many existing measures of women’s empowerment have

ambiguous associations with indicators of nutritional status. This is likely because existing

operationalizations of empowerment often focus on aspects that are somewhat distal from factors that

influence nutrition. In this paper, we present an index that aims to measure women’s empowerment in

the realm of nutrition. We define nutritional empowerment as the process by which individuals acquire

the capacity to be well fed and healthy, in a context where this capacity was previously denied to them.

Our index draws on theory and multi-site formative research from South Asia and captures multiple

dimensions of empowerment spanning domains that influence nutritional outcomes. We construct this

index using data from two sites in India and validate it by estimating two nutritional outcomes, body

mass index (BMI) and anaemia, as a function of the index. We find that our index is significantly

associated with these outcomes, indicating that in rural South Asia, the women’s empowerment in

nutrition index can assist researchers to understand the nutritional status of women and their families.

Keywords: gender, rural, nutrition, South Asia, empowerment

JEL Code: J16, D63, I00

1

The Women’s Empowerment in Nutrition Index1

Sudha Narayanan, Erin Lentz, Marzia Fontana, Anuradha De, and Bharati Kulkarni

2

Abstract

Rural women face many obstacles that thwart their well-being. Policies that seek to empower them, for

example, by improving livelihood opportunities, often do not translate into improvements in other areas,

notably in their nutritional status. Indeed, many existing measures of women’s empowerment have

ambiguous associations with indicators of nutritional status. This is likely because existing

operationalizations of empowerment often focus on aspects that are somewhat distal from factors that

influence nutrition. In this paper, we present an index that aims to measure women’s empowerment in the

realm of nutrition. We define nutritional empowerment as the process by which individuals acquire the

capacity to be well fed and healthy, in a context where this capacity was previously denied to them. Our

index draws on theory and multi-site formative research from South Asia and captures multiple

dimensions of empowerment spanning domains that influence nutritional outcomes. We construct this

index using data from two sites in India and validate it by estimating two nutritional outcomes, body mass

index (BMI) and anaemia, as a function of the index. We find that our index is significantly associated

with these outcomes, indicating that in rural South Asia, the women’s empowerment in nutrition index

can assist researchers to understand the nutritional status of women and their families.

Keywords: gender, rural, nutrition, South Asia, empowerment

JEL Codes: J16, D63, I00

1 This paper is part of a project titled Women’s Empowerment in Nutrition Index (WENI) funded by a grant from

Innovative Methods and Metrics for Agriculture and Nutrition (IMMANA) of DFID. We thank Aastha Ahuja, Mohit

Sharma Padmaja Pati, Parul Saboo, Shree Saha, Sweta Bhushan, Udayan Rathore and our survey teams for their

assistance and Gal Reddy, for crucial technical support in the field. We also thank Chris Barrett, Chris Oldiges and

participants at a 2019 WENI workshop in Delhi for their valuable feedback. We alone remain responsible for any

errors and omissions that remain. 2 Sudha Narayanan is Associate Professor, Indira Gandhi Institute of Development Research (IGIDR), Mumbai,

India; Erin Lentz is Assistant Professor at the Lyndon B Johnson School of Public Affairs, University of Texas,

Austin, USA; Marzia Fontana is Research Associate at the Institute of Development Studies at Sussex, UK;

Anuradha De is Director, Collaborative Research and Dissemination, New Delhi, India; Bharati Kulkarni is Senior

Grade Deputy Director at the National Institute of Nutrition, Hyderabad, India.

2

The Women’s Empowerment in Nutrition Index

1. Introduction

Poor rural women’s lives are characterized – more so than rural men’s – by a multiplicity of tasks

and responsibilities. This is a reflection and consequence of “women’s work” including both productive

and reproductive (care) activities while men’s work tends to focus primarily on production (FAO-IFAD-

ILO 2010). As noted in a UN World Survey on the Role of Women in Development (UN Women 2014),

governments and donors increasingly target women as important agents for promoting sustainable food

production, enhancing household food security and protecting the environment. But if projects are

designed without due consideration to the interdependencies and conflicts between women’s productive

and reproductive roles, the norms that govern these roles and the consequences for women’s overall work

burden, gender inequalities may intensify. Further, even projects leading to empowering experiences in

one area of a woman’s life (e.g., a greater say in agricultural production decisions or the opportunities to

do paid work) do not automatically translate into her greater capacity to make choices and transform

power relations in another area of her life (e.g. entitlement to an adequate and nutritious diet).

This potential disjunction is particularly important in the context of current efforts that aim to

empower women economically on the one hand and secure their nutritional wellbeing on the other (UN

Women, 2018). In fact, studies that measure empowerment in different areas suggest that the linkage

between economic empowerment and nutritional wellbeing are complex, particularly in rural contexts

(Kennedy, 1994; Kadiyala et al. 2014). For example, recent analyses aiming to establish the relationship

between women’s empowerment in agriculture (WEAI) and nutrition-related variables (such as per capita

calorie availability, dietary diversity and adult BMI) find mixed evidence in both West Africa and South

Asia (Malapit and Quisumbing 2014 for Ghana; Sraboni et al. 2014 for Bangladesh; Malapit et al. 2015

for Nepal). Such findings highlight the need for a complementary set of measures and indicators that

more explicitly captures the pathways leading to women’s empowerment in nutrition. Thus, if improving

3

nutritional wellbeing and the empowerment of women are both goals, it is crucial not only to identify and

address the constraints women face in securing nutrition but also to determine how women can attain

better nutritional outcomes through processes that support the expansion of meaningful choice for women

(Kabeer 1999).

To this end, we present an index, the Women’s Empowerment in Nutrition Index (WENI), that

aims to measure empowerment in nutrition. We define nutritional empowerment as the process by which

individuals acquire the capacity to be well fed and healthy, in a context where this capacity was

previously denied to them (Narayanan et al. 2017). This process entails acquiring knowledge about, and a

say over, nutritional and health practices; gaining access to and control over intake of adequate and

nutritious food; and being able to draw support from both family and other institutions to secure and

maintain an adequate diet and health.

The WENI is a multidimensional index that draws on and brings together two streams of

literature – drivers of nutrition and empowerment – and multi-site formative research from rural South

Asia. We apply this index to data from two sites in India to understand the role of empowerment in

nutritional outcomes – both to assess the extent to which individuals are empowered but also to identify

those who are disempowered. The paper demonstrates the use of this index to generate measures of

disempowerment at the individual and community levels that help understand obstacles in attaining

desirable nutritional status.

Our index differs from other empowerment measures in several conceptual ways. First, it focuses

on empowerment in nutrition – an outcome – rather than livelihoods defined around agriculture/livestock

(e.g., Alkire et al. 2013; Galie et al. 2018). Second, it is concerned about women’s empowerment in

nutrition in general, including the nutritional empowerment of rural women whose households are not

engaged in agriculture, who belong to landless households, whose families rely on common pool

resources, and/ or who rely on remittances. It is common for rural households to pursue pluri-activities

4

and in many rural communities, links to agriculture are indirect (e.g., coastal, forest and pastoral

communities and those who have caste-based occupations). Further, our approach does not assume

empowerment in one area (e.g., empowerment relative to production decisions) will translate to another

(e.g., nutrition) and recognizes that empowerment is a complex, multifactoral process (Kabeer 1999;

Robeyns 2003). We attempt to capture different aspects of empowerment, including agency, resources,

and achievements. Many empowerment measures tend to focus on agency, with less emphasis on other

aspects (Donald, et al, 2017; Meinzen-Dick, 2018). However, in the context of nutrition, having agency is

not enough to guarantee nutritional empowerment. Knowledge and broad institutional resources are

necessary for women to achieve their adequate nutritional status. For example, Lentz et al. (under review)

find that in rural Odisha, women describe advocating for access to social protection schemes to which

they are entitled. These schemes could support nutritional outcomes for themselves and their families.

Yet, local authorities remained unresponsive, contributing to women’s food and nutritional insecurity and

disempowerment.. Our effort is therefore to explicitly incorporate relevant resources as well as norms

shaping the terms of access to such resources, (Kabeer 1999; Robeyns 2003; Alkire et al. 2013).

Finally, we also shift the focus away from children to women and other household members. Literature on

child malnutrition strongly suggests that women’s empowerment influences children’s nutritional status

and that child nutritional status itself is often positively associated with maternal BMI (Pratley 2016). Yet,

it is not clear what aspects of women’s empowerment matter for women’s own nutritional status; or

whether men or women (both those in the reproductive age group and those who are not) face different

barriers. We hope that the WENI offers a way to address some of these issues.

Our paper is organized in eight sections. In Section 2, we elaborate on the concept of nutritional

empowerment. We then discuss the rationale for an index to assess nutritional empowerment and discuss

the steps involved in constructing the WENI (Section 3). After a brief description of the survey that forms

the basis for our illustration of the WENI (Section 4), we discuss the findings (Section 5). In Section 6,

we assess the ability of the WENI to predict nutritional outcomes. We also examine the sensitivity to

5

alternative assumptions and approaches to measurement to ensure its robustness (Section 7). Section 8

concludes the paper with a discussion on how to and how not to use the WENI and possible future work.

2. Conceptualizing nutritional empowerment

We believe women’s nutritional statuses matter in their own right and our index of nutritional

empowerment focuses on women themselves, rather than their children.3 While our effort aims to

measure women’s empowerment in nutrition (WEN), the concept of nutritional empowerment has been

developed more generally for any individual. Our particular interest is to examine revealed gender

disparity in nutritional empowerment via comparisons between men and women, or among women of

different age groups.

We identify four dimensions of nutritional empowerment. We begin with Naila Kabeer’s (1999)

dimensions of empowerment: resources, agency, and achievements. We also single out knowledge as

distinct fourth dimension. Because we are interested in better tracing the barriers women face to making

strategic life choices, we aim to disentangle lack of knowledge about material and social resources and

how best to use these versus actual lack of such resources. Within resources, we consider material and

social resources as well as the terms and norms that define access to them. We understand agency as

including intrinsic, individual and collective agency (Kabeer, 1999). We define achievements as

nutritional outcomes. Knowledge, resources and agency constitute the dimensions of empowerment we

seek to capture, while achievements are the nutritional outcomes we care about.

We consider three main domains as particularly relevant for nutritional empowerment – food,

health, and institutions. These are consistent with the UNICEF approach (1999) to understanding child

malnutrition, which is relevant also for adult malnutrition. Our domain of institutions encompasses

3 While women might be well fed and healthy without their agency having led to it, for example, where food and

other material resources are abundant and easily available, we are especially concerned about situations of scarcity

and deprivation. In these situations, a woman’s knowledge and capacity to act to secure food might make a

difference to her nutritional status, particularly in the face of strong discriminatory gender norms. It therefore

matters if nutritional outcomes reflect a meaningful choice or not.

6

UNICEF’s underlying causes; food and health domains capture UNICEF’s intermediate causes. We

exclude childcare practices, which are less relevant for able-bodied adults. Across each domain, we

consider knowledge, resources and agency. We also highlight paid and unpaid work as a crosscutting

factor that feeds into both food and health domains. Work provides either direct access to food (as with

those engaged in cultivation and animal husbandry) or resources that can help support adequate and

quality diets. Work also entails energy expenditure and potential exposure to health hazards. Therefore,

quality of work is also a crucial health input, with poor working conditions becoming a “negative”

resource. Including paid and unpaid work, rather than focussing on a specific livelihood (e.g., agriculture

or livestock) expands the usability of our index, making it relevant for all rural populations rather than

those involved in a specific form of work. We introduce another sub-domain, fertility, which is potentially

a critical aspect of nutritional empowerment for women of reproductive age (between 15 and 49)..

Findings and comparisons regarding the role of fertility in nutritional empowerment are limited to women

of reproductive age. We capture the norms and cultural practices that shape food and health domains,

such as the order of eating within a household, within those domains,. However, we also include a

separate domain of broader institutions, referring to legal rules, legal pluralism, and societal and

community norms likely to indirectly influence nutritional empowerment.

The WEN Grid (Figure 1) reflects each of these domains and dimensions and . forms the basis

for operationalizing our survey (Table 1). For an individual there are indicators associated with seven

domain-dimensions of nutritional empowerment and for women of reproductive age there are ten domain-

dimensions. Each D-D can be scored based on the proportion of D-D-specific indicators in which an

individual is deemed to be empowered, which yields a nutritional empowerment scorecard for an

individual. The scorecard helps identify the D-Ds where there are most obstacles at both individual and

community levels. Aggregating these D-D scores into the WEN index provides us with a comprehensive

measure to gauge progress of communities with respect to nutritional empowerment.

7

Figure 1: The WEN Grid and its domain-dimensions

Why an Index

Operationalizing a measure of nutritional empowerment that correlates with nutritional status involves

two formidable challenges. First, measuring empowerment is far from straightforward (Kabeer 1999;

Robeyns 2003; Gates 2014). As Kabeer (1999) has argued, some indicators that are intended to capture

empowerment fail to do so. There is also a view that it is not always desirable or easy to quantify what is

essentially a complex process. At the same time, a rich tradition of empirical quantitative work has

generated as many as 121 unique indicators of empowerment in the context of studying its influence on

nutritional outcomes (Pratley 2016). Most of these measures of empowerment are confined to

representing aspects for which secondary data are available and do not specifically capture empowerment

in the realm of nutrition.

BMI, no anemia

Food adequacy

MDDS-W

CSI

Free from chronic illnesses, morbidities

Adequate Birth spacing,

No Child mortality

Final achievement

Intermediate achievement

Resource Agency Knowledge

Institutions

Fertility

Health

Food

Work

Work

8

A second challenge is the wide range of factors that collectively influence nutritional status. The

effort of identifying precise relationships between empowerment and nutritional outcomes becomes

complicated very quickly. In such situations, indices can be valuable tools to understand concepts that are

complex and multidimensional. Despite their recognized limitations, notably the loss of detail (Meinzen-

Dick 2013), and questions of how to aggregate and weight factors (Alkire and Foster 2011; Alkire et al.

2012), indices remain useful ways to aggregate and communicate findings to initiate policy dialogue and

discussion.

Notwithstanding the challenges inherent in measuring empowerment, we believe that the

conceptualization of nutritional empowerment itself and the attempt to measure it can (a) clarify our ways

of thinking about the socio-economic contexts that shape nutrition outcomes, differently or not, for

women and men, (b) serve as a basis or tool for diagnosing the weakest domain/dimensions of nutritional

disempowerment via the scorecard and decompositions of the index itself (described later), and (c)

(hence) support efforts of policy makers to address critical gaps in order to promote nutritional security

and health, specifically of women. In what follows, we describe how to construct the index, identify

relevant factors and compute aggregate sample-level indicators of nutritional disempowerment in the

community.

3. Constructing the Index

Our approach is to identify indicators to quantify elements for each D-D in the WEN grid (Table 1). We

then score each domain-dimension to gauge the fraction of total indicators in which an individual is

empowered within the D-D, to obtain a scorecard of nutritional empowerment. This scorecard is the basis

for constructing an index but is itself of value to identify D-Ds where an individual is least empowered.

We then aggregate these scores into an index. The primary purpose of the WENI is to gauge the extent of

empowerment specifically in the domain of nutrition; but the WENI is also constructed to be predictive of

nutritional status, and we validate this feature of WENI later. The predictive power of the index is only

9

true on average, i.e. it isn’t true that every individual who is more empowered will also have a higher

nutritional status.

To construct the WENI, we draw on literature on the construction of Multidimensional Poverty

Indices (MPI) (Alkire et al. 2015). As with the MPI, the central tasks of constructing the WENI involve:

(1) defining a set of relevant indicators, (2) defining an empowerment cut-off normatively for each

indicator; individuals below that value are disempowered for that indicator (3) creating new binary

indicators that assigns individuals above the cutoff a value of 1 (indicating being empowered in that

indicator) and assigns a 0 to those below the cutoff, (4) assigning a weight to each new binary indicator,

and computing empowerment scores for each D-D, (5) constructing an empowerment index by

aggregating across the scores (weighted or not) for each D-D, and (6) setting a threshold for the overall

index such that if the person has a score below the threshold, the person is considered nutritionally

disempowered.4 (7) compute the proportion of the sample who are disempowered and the intensity of

disempowerment.

Several steps involve normative judgements, especially the choice of weighting systems and the

threshold (Alkire et al. 2015). Our thresholds represent aspirational levels but from a policy perspective

could be set to pragmatic goals attainable in the short/medium term.

The MPI approach to index construction offers several aspects useful for policymaking and

evaluation. First, the MPI approach satisfies subgroup decomposability (Alkire, et al. 2015). In the

WENI, subgroup decomposability ensures that the empowerment level in a society is equivalent to the

population-weighted sum of subgroup empowerment levels. Further, the MPI approach also allows us to

break down the WENI by domain and dimension. This offers a useful policy diagnostic tool to identify

4 In the typical MPI, deprivations are counted to identify those that are multidimensionally poor/deprived. In the

WENI we count the achievements. While the approach is different they are equivalent in the sense that for MPI the

minimum cut-off in the MPI is the threshold of deprivations and in the case of WENI, the threshold we use,

represents the maximum achievements so that those below the threshold are regarded as nutritionally disempowered.

10

disempowerment by dimension and or domain of empowerment. This approach can be combined with the

D-D specific scores to examine where an individual, community, or target group fares worst. This use of

WEN constituent indicators as a diagnostic comprises the “scorecard approach” and can be applied even

without aggregating the scores into the WENI (discussed further in Narayanan et al. 2017). The WENI

measures also satisfy dimensional monotonicity. When an individual moves from being disempowered to

empowered in particular dimensions or domains, overall empowerment levels on the aggregate increases.

This is quite useful for tracking empowerment as a process, especially given that empowerment can occur

in different domains at different rates (Kabeer 1999). It would also help uncover the interdependence of

the domains and dimensions such that an intervention in one domain-dimension might empower the

beneficiary in another domain or dimension.

The index aims to capture enduring changes in nutritional empowerment rather than temporary

fluctuations, although it accounts for seasonal variations that can contribute to nutritional empowerment.

For example, one potential indicator captures whether paid work opportunities are available throughout

the year. If such opportunities for continuous employment emerged and replaced seasonal work, the index

would capture this positive change.

3.1. Identifying Relevant Indicators

The first four steps of computing the WEN index include: identifying which indicators are

relevant, determining indicator-specific empowerment cutoffs, generating binary indicators from these

variables, and aggregating these within the D-D by applying a set of weights.

. The survey we use to illustrate WENI (henceforth the WENI Survey, described below) includes

questions on each D-D and on ultimate achievements. Our measures of achievements are haemoglobin

levels and BMI, both indicators of nutritional status for adults; other measures may be appropriate in

other contexts. Table 1 shows the main themes and several candidate indicators could populate each

domain-dimension. For each indicator or theme, our survey included multiple approaches to capturing

11

the indicator. How one measures each indicator and which questions from a cluster of potential questions best

captures a theme could vary by context. For example, to capture norms around eating, we ask questions ascertaining

whether the family members eat together. Another alternative is to ask if the individual has the freedom to choose

when to eat or to ask how frequently the individual eats last.

Our indicators are thus derived from a cluster of survey questions, each potentially a good

construct of the theme we wish to capture. However, we do not prescribe specific questions, which would

depend on the context and the sub-population studied. In the selection of questions, there are a few

principles useful to consider. We took care to ensure that some indicators are free from gender

specificities, so that the question asked is as relevant for men, as it is for women of different age groups.

As with any index, it is challenging to use and interpret indicators reflecting processes that may generate

multiple effects with opposite signs. For example, doing paid work can facilitate nutritional

empowerment as a source of income but can undermine it when it involves disproportionate energy

expenditure. We made an attempt to capture all of the different aspects of such variables. Further, several

indicators can be measured at either the individual or household level. To reduce the survey burden,

several indicators were measured at the household level (i.e., access to toilets, and drinking water),

although we acknowledge that measures at the individual level will better capture the drivers of intra-

household variation in nutritional empowerment. Other indicators are subjective assessments and as such

inherit the problems associated with most subjective measures. For example, rather than an elaborate

time-use survey we opted to get the respondent’s perception of their leisure time, in order to reduce

survey burden. These remain caveats. We note also that it is simpler to construct the index from questions

where conflicting or inconsistent responses are less likely; pre-testing survey questions and adequate

enumerator training can reduce this issue. For more information on selecting questions to generate the

indicators, see the supplementary appendix.

From the rich set of variables collected as part of the WENI Survey, where appropriate, we use

the variable as asked while , in other cases, we combine survey questions to generate indicators. We

12

convert each indicator in each D-D into a binary variable. For binary variables, 1 represents

empowerment and 0 represents disempowerment. For continuous and ordered variables (i.e., Likert-

scale), higher values indicate greater indicator-level empowerment. For simplicity, we define an

indicator-specific threshold of 0.5, for all continuous indicators, such that a person scoring above that

value for a specific question is deemed to be empowered (and assigned 1) when transforming these into

binary indicators. Policymakers and practitioners with interest in specific indicators may wish to adjust

thresholds (Alkire et al. 2015). The complete dataset from the WENI Survey includes 128 indicators

derived from 281 variables. When including the fertility domain for WEN women, there are 158

indicators derived from 356 variables. We denote this full set of potential WEN Grid indicators as F.

One issue common to most indices is that each respondent must have an associated value for each

indicator in order to be included in the computation of the index. Yet, as is the case with many indices,

some WEN survey questions are not applicable to all respondents. We use normative judgements to code

these “Not Applicable” (NA) values as empowering or disempowering to retain these individuals in our

index. For example, questions regarding conditions of paid work are applicable to only those who access

paid work. For an indicator on energy intensity of work, we code “NA” as empowering because not doing

paid work is equivalent to saying that there was no energy expenditure on paid work. This sometimes has

the effect of masking significant differences in responses across individuals, and is a limitation of indices

more broadly. The score relative to energy expenditure of a woman not doing paid work could be still low

(i.e. ‘disempowering’) if her not doing paid work is more than offset by long hours , or intensive, unpaid

domestic work” or similar. In the current version we are able to address this in a limited way, treating

support for domestic and care work as a positive resource. Table 2 provides values for key indicators, and

indicators how NA responses influence such values.

From the full set of potential empowerment indicators, F, we identify a shorter list of indicators,

S, that feeds into WENI. Within each domain-dimension, we first use factor analysis, a data reduction

technique that aims to identify latent factors related to a larger number of variables. The value of factor

13

analysis is that it helps identify which subset of indicators to include, such that the indicators are not

correlated with one another. It also helps select specific formulations and constructs of indicators that

“best” represents each D-D. We conduct a factor analysis separately for WENI woman, spouse and

mother in law and within each D-D. We choose those variables whose eigenvalues (a measure of total

variation accounted for by each factor) are over 2, with factor loadings over 0.4 (Acock, 2011).5 We call

this smaller set of indicators S; S is a subset of F. As a consistency check, we compute the

tetrachoric/polychoric correlations of elements of F with those variables that were discarded (i.e., that are

part of F but not of S) to ensure that the factor analysis did not drop variables that might be uncorrelated

with those in S and might be relevant. In the section on sensitivity analysis and validation, we discuss

results using other factor loadings. In each D-D, ten or less indicators met the factor analysis selection

criterion. A benefit of having a maximum of ten indicators in each D-D is that it helps to ensure that

WENI is parsimonious and tractable.

We then apply a normative lens to the indicators identified through factor analysis, the set S, to ensure

the WENI stays anchored to our conceptualization of nutritional empowerment. This involves grouping

some variables, checking for redundancies, and including some indicators that factor analysis might have

not captured. One example is water and sanitation. While the factor analysis results suggest that these

indicators do not appear relevant in the context of our study, we include them nevertheless to reflect

evidence that WASH practices and resources often matter for nutrition in many contexts (Headey et al.

2015). We denote this final subset of indicators as I. I is the basis of WENI and contains 40 indicators

derived. Including the fertility domain, the total number of indicators is 50. Table 2 lists the specific

measures included in I. The indicators within each D-D in Table 2 represent a substantive theme or aspect

of empowerment that might matter for nutrition.

3.2. Identification of disempowerment

5 Costello and Osborne (2015) recommend 0.3.

14

Having identified the set of relevant indicators, we then compute the score for each D-D. Each score is the

ratio of indicators an individual is empowered in to the total number of indicators within that D-D. This

set of D-D specific scores allows us to generate a scorecard of nutritional empowerment based on the

WEN Grid. The WENI builds on this scorecard using a counting approach. For each D-D we establish

thresholds below which people are considered nutritionally disempowered. We set the cut off for each

domain-dimension at 0.5. If there are n indicators representing a domain-dimension, then an individual is

considered empowered in that D-D if the individual scores at least n/2 for that D-D.

We then aggregate these D-D-specific empowerment indicators over the 7 domain-dimensions

(10 for WENI women of reproductive age). A simple count of the number of domain-dimensions in

which an individual is empowered is divided by the total number of domain-dimensions (i.e., 7 or 10) to

obtain WENI. In computing the WENI, we assign equal weights to each domain-dimension. We do so

because we believe that empowerment involves having knowledge, access to resources and supportive

structural conditions as well as agency to act upon the knowledge and utilize the resources, across all the

domains. The WENI ranges from 0 to 1. A person who is completely nutritionally disempowered scores 0

and someone who is fully empowered scores 1. The former is disempowered in all 7 (or 10) domain-

dimensions and the latter is empowered in all. An individual who is nutritionally empowered is thus

empowered in each domain-dimension, which means that (s)he is empowered in at least half the domain-

dimension-specific indicators for each domain-dimension.

The implication of first weighting D-D specific indicators equally within the D-D and then

aggregating over seven (or ten) D-Ds again with equal weights is that each indicator in D-Ds with fewer

indicators get a disproportionately higher weight. For example, if there are three indicators for Food-

Knowledge, each indicator is given 1/3 weight for the D-D specific score. In the aggregate WENI, each

Food-Knowledge indicator gets a weight of 1/21 because each of the 7 domain-dimensions get a weight

of 1/7. Food-Resource, on the other hand, has 5 indicators, implying that each Food-Resource indicator

15

would hence get a weight of 1/35. A benefit of this approach is that it ensures that the ranking is not

driven overwhelmingly by the number of indicators in each domain-dimensions.

We then use a cut-off of 0.5 on the total nutritional empowerment index. Those with a WENI

below 0.5 are considered nutritionally disempowered. For the WENI to be a reliable measure of

nutritional empowerment, it should rank individuals similarly whether one uses F, S or I. In the section on

sensitivity, we show that this is the case and the rank order is maintained.

Many indices, such as the WEAI (Alkire et al. 2013), include gender parity as a component of the

index. We do not. Gender parity cannot often be reduced to spousal parity in the context of complex

family structures and it is challenging in a household survey to capture the disparities between men and

women in a meaningful way. We also believe that in resource-constrained settings, where deprivations are

pervasive, it is not clear what gender parity implies when both men and the women are deeply - if not

equally - deprived. We therefore restrict ourselves to measures of individual empowerment and a

comparison of WENI across these two groups is the only sense in which we can capture or understand

intra-household disparities.

4. The Survey

We use a fit-for-purpose survey (henceforth, the WENI survey) that measures the variables represented in

each domain-dimension in the WEN Grid. The WENI survey resamples women from 26 villages in

Odisha and Bihar who were previously surveyed under the 2014-2016 Systematic for Promoting

Appropriate National Dynamism for Agriculture and Nutrition (SPANDAN) project. See the

supplementary appendix for more information on sampling. By resampling SPANDAN interviewees, we

can draw on SPANDAN’s baseline demographic information (e.g., caste and prior fertility information),

decreasing the survey burden for our respondents. Women in our sample had at least one child under age

seven, and are identified as WENI women. Among households with a husband present, every other

husband of WENI women was interviewed. The same sampling procedure was used for households with

16

mothers-in-law, but in general there were fewer mothers-in-law than husbands. Our sample included 522

WENI women, 290 husbands and 161 mothers in law. We caution that this sample is not representative at

administrative levels higher than the village level and is representative only of households with a child

below the age of 7 years. We do not report village names to maintain respondent anonymity. Thus, the

purpose of the survey is to operationalize and validate the concept of nutritional empowerment.

The survey was conducted in February to April 2018, using two teams of trained enumerators.

When possible, interviews with WENI women, mothers-in-law, and spouses were conducted

simultaneously and in private, to enable WENI women to speak frankly.6 The Institutional Review Board

of National Institute of Nutrition, India reviewed survey protocols. Enumerators received training on

ethical research practices and requesting informed consent. Survey length for the WENI woman was

about 75 minutes; for other household members it was about 40 minutes. As compensation for the time

the household spent on the survey, we offered a kilogram of cooking oil, valued at Rs. 100.

The survey collected information on each domain-dimension of the WEN index for each

respondent in order to compute her or his WEN score. We obtained the weight and height for each

interviewee to compute body mass index (BMI); low BMIs in adults indicate undernutrition. We also

collected dried blood spot (DBS) samples from the WENI women to test hemoglobin (Hb) levels; low Hb

indicates anemia, a micronutrient deficiency. BMI and anemia are the ultimate achievements against

which we internally validate the index. Additional details on the survey and on steps taken with

respondents with low hemoglobin levels are in the supplementary materials.

Within the sample, we find rates of undernutrition and anemia are consistent with levels in the

respective states as per the National Family Health Survey 2015-16 (Table 3). A quarter of our sample

was underweight; 65% had mild anemia or worse and 39% had moderate anemia or worse. About 47%

6 56 of 454 WENI women were interviewed in presence of someone. For 68 cases, we do not have this information.

17

reported that they had been sick the month prior to the survey and 22% suffered from one or more chronic

illnesses. There are stark gender differences; on average, women fare worse than men.

5. Results

We first use WENI to identify which communities or social categories are nutritionally disempowered

overall. We then consider the specific domains and dimensions where individuals are systematically

disempowered, decomposing the WENI itself and using the scorecard of the WEN Grid to complement

the analysis.

5.1 Who is disempowered?

We first calculate the prevalence of nutritional disempowerment equivalent to the Headcount

Ratio H. H is the proportion of those nutritionally disempowered divided by the entire sample. We also

compute the intensity of nutritional disempowerment for the entire sample, which is the percentage of

deprivations suffered by each individual (i.e., the percent of domain-dimensions where an individual is

disempowered). We also present these by sub-groups - mothers-in-law, spouses, and WENI women, by

state.

In many resource-constrained contexts, some households might be systematically more intensely

disempowered than others, which is the depth of disempowerment, G,. From a policy perspective in

particular, it is useful to identify how far below the cutoff those who are nutritionally disempowered are.

This is computed as ∑ | ̅|

.

We then generate a set of measures ( ) that sheds light on the nature and extent of nutritional

disempowerment for communities. The adjusted headcount ratio ( ) combines the head count and

intensity of nutritional disempowerment, computed as . It allows us to understand how, even

with the same headcount ratio, communities might still differ in the intensity of nutritional

disempowerment. The adjusted gap ( ), measures the depth of the extent of nutritional

disempowerment, which captures how far below the cut off a community is on average. This measure

18

combines incidence, intensity and depth and is computed as . The higher the value of

each of these metrics, the greater the nutritional disempowerment.

Distributions of the Index appear comparable across states (Figure 2) but are significantly

different for WENI women, their spouses and mothers-in-law. Overall, as expected, the distribution of the

index for male spouses is centred right of that for the WENI women. Mothers-in-law, on average, have

lower scores than for either of the other two groups (Figure 2). Stochastic dominance tests show that the

distribution of WENI for male spouses first order dominates those of WENI women and their Mothers-in-

law.7 Kolomogorov-Smirnov tests for equality of distributions of WENI too suggest that WENI scores

male spouses is statistically significantly different from those for WENI women.

Using the WENI cut-off of 0.5, we find that 66% of individuals in the sample are nutritionally

disempowered. A gender-wise comparison suggests a stark difference between men and women (Figure 2

and Figure 3). Both overall and within each state, twice as many women are disempowered compared to

men. The gap is larger in Odisha than for Bihar. In Odisha, about 29 percent of male spouses are

disempowered, but 79 percent of WEN women are. The rates of nutritionally disempowered mothers-in-

law are worse than the rates for WEN women. Among mothers-in-law, 89% in Odisha and 81% in Bihar

are nutritionally disempowered, compared with 79% and 75% respectively for the WENI women. In the

estimations for WEN women that includes fertility, we also find that differences in the headcount ratio

and intensity of nutritional disempowerment between Bihar and Odisha are small.8

The difference between genders is starker when one factors in the intensity of disempowerment

and the disempowerment gap. Whereas the headcount ratio suggests women are a little more than twice

as likely to be disempowered as men, the adjusted headcount ratio that takes into account intensity of

disempowerment, suggests that the WEN women are three times as more likely to be disempowered

7 The relationship between WENI women and their mothers-in-law is indeterminate – there is neither

first order nor second order dominance. 8 We note that while the intensity of disempowerment is lower for WENI women when the fertility domain is

included compared to results from the pooled approach. However, these scores are not directly comparable because

two different sets of indicators were used to generate the scores.

19

(Table 4). In other words, women are more likely than men not only to face obstacles in more domain-

dimensions but also to fall short of D-D specific thresholds. Nutritionally disempowered women thus

have a longer distance to “catch-up” than do nutritionally disempowered men.

While many men are more nutritionally empowered than women, they too continue to face

obstacles on many domains and dimensions. In some cases, they are far below the threshold. Thus, gender

disparity in empowerment prevails in contexts of more general and widespread deprivation.

Figure 2: WENI in Bihar and Odisha

Figure 3: Distribution of WENI by WENI Women, Spouses and Mothers-in-law

20

Note: The Kolomogorov-Smirnov statistic for equality of distribution of male spouse vis-à-vis WENI

women is 0.4887 and MIL is 0.6148, both statistically significant at the 1% level.

5.2 Which domains and dimensions require policy attention?

Because of its decomposability, WENI can support identification of where the rates of

disempowerment are the highest. The WEN scorecard helps identify which aspects of nutritional

empowerment are weakest for which groups and which D-Ds most contribute to average nutritional

empowerment scores (Figure 4 and Figure 5). Most of the differences in the nutritional empowerment

index between men and WENI women reflect differences in the health agency D-D and institutions

domain (Figure 4). Health agency includes indicators such as the ability to go alone to a health center, and

the ability to make a decision alone about one’s own health (Table 2). The institutions domain includes

indicators for freedom of movement, participation in the public sphere, access to information. In almost

all of the component indicators in the institutions domain, men fare way better than the either WEN

women or mothers-in-law. For example, men are more likely to be in control over their own health

decisions and can visit the medical facility alone without having to seek permission (See Table 2).

The two groups of women in our sample also offer interesting contrasts. The difference between

WENI women and mothers-in-law comes additionally from the dimension of knowledge, be it food or

health. Younger women seem to have more knowledge than their mothers-in-law and perhaps also of

agency, notably in the domain of health. Yet, mothers-in-law have greater health resources at their

disposal. They are able to get assistance in household chores if they are ill, and they are more likely to be

satisfied with the level of assistance they get regarding chores (Figure 4 and Table 5).

If one wants to understand which domain-dimensions are weak and are the obstacles to better

nutritional empowerment, the scorecard is useful (Figure 4; Table 5). The low mean empowerment scores

for the knowledge dimensions suggest that knowledge may be a particular barrier to nutritional

empowerment. Across states and groups surveyed, the scores are fairly low, particularly for mothers-in-

law. Regarding work, we find that men do substantially more paid work and disproportionately report

poorer working conditions and backbreaking work. Women in contrast report more domestic and care

21

work and getting less support in undertaking such work (Table 2). For program evaluation purposes,

baseline and endline WEN scores could show how changes in specific indicators or specific dimension-

domains influence nutritional empowerment.

Figure 4: Contribution of Each Domain-Dimensions to the Nutritional Empowerment Index

Figure 5: A Scorecard for WEN Domain-Dimensions

22

6. Validation

We now examine the WENI’s prediction of nutritional outcomes. While we care about nutritional

empowerment in its own right, we also care about nutritional achievements. Given the indicators’

normative basis for inclusion, not all will necessarily have a predictive relationship with nutritional

achievements. Thus, we aim to ensure that the indicators, I, that constitute WENI are collectively able to

explain a significant proportion of the variability in BMI and anemia.

Figure 6 graphs WENI against BMI; WENI is positively correlated with BMI for the entire

sample and for most subgroups, suggesting that higher nutritional empowerment is plausibly associated

with better BMI.9 A similar pattern is apparent for haemoglobin levels (Figure 7). Table 6 presents results

for t-tests of equality of mean nutritional outcomes and of equality of proportions with adequate

nutritional status by empowerment status. In both cases, we reject the null of no difference. As a more

robust approach, we estimate a series of regression equations to evaluate the extent of variation in BMI or

haemoglobin levels that the WENI explains. We estimate the following linear models:

( )

( 10 )

Our outcome variables are the continuous (and binary) nutritional outcome measures of BMI and

haemoglobin levels for individual i. We estimate least squares models for both, i.e. linear probability

models for the binary outcome variables, with robust standard errors (Table 7). We note that BMI should

9 Since the sub-groups are fairly small, the results at the sub-group level needs to be interpreted with caution. We

only present results for which we have a large sample.

10 We use the World Health Organization (WHO)’s thresholds for specific age groups and gender and physiological

status (pregnant/nursing). For men it is 130 g/l, pregnant women, it is 110 g/l, non-pregnant women 120 g/l

23

not be interpreted for pregnant women. and exclude them from all models. We exclude household and

individual level controls, because systematic differences in household socio-economic status are captured

adequately, if not fully, by the WEN Grid. However, including a vector of controls such as house type,

roofing material, number of rooms in the house, religion, caste, age, and household wealth index of the

respondent does not alter the results (these are available with the authors).

Figure 6: The Relationship between WENI and BMI

Note: These are local polynomial regressions; confidence intervals have not been plotted here.

24

Figure 7: The Relationship between WENI and Haemoglobin levels

Note: These are local polynomial regressions; confidence intervals have not been plotted here.

7. Sensitivity analysis

First, we apply sensitivity checks regarding indicator selection. As described above, the indicators

in the different domain-dimensions of nutritional empowerment have a normative basis for inclusion.

While the complete set of indicators is most informative, several indicators are highly correlated with one

another and the need to collect many indicators can be a barrier to the uptake of WENI. For these two

reasons, the final subset of indicators, I, were derived using factor analysis. There is not a definitive

factor loading cut-off for identifying a subset of indicators for a given domain-dimension factor. For this

reason, we test an alternative factor-loading cutoff of at least 0.63, based on recommendations made by

Tabachnick and Fidell (2007). Variables were included based on this cutoff conditional on their

eigenvalues being greater than 1 and cumulative variance adding to unity. At the aggregate level, the

WENI slightly changes and the rank order is preserved.

25

Second, to evaluate the sensitive of the WENI to indicator selection, we vary indicators within

domain-dimensions. Our survey collected a variety of similar indicators. Each indicator could be an

equally appropriate or suitable indicator from a normative perspective. We select one indicator from

among a group of highly correlated indicators within a D-D, and compute WENI; we then compute

WENI for a different indicator from that same group of correlated indicators. Here too, the WENI seems

to vary slightly but the rank order is highly correlated across the alternatives.11

Third, we assess the sensitivity of the empowerment index to selection of the threshold above

which individuals are considered nutritionally empowered. In this paper we used a threshold of 0.5.

Without applications of WEN in other contexts, we do not yet know whether identifying universal

thresholds is appropriate. In resource-constrained settings, too high a threshold renders everyone

disempowered. A threshold too low would make everyone empowered – making measures such as the

headcount ratio and comparisons between groups or communities on that basis meaningless. One

recommendation is to set the threshold to reflect aspirations, for example, by drawing on policy or

programmatic goals (Alkire, et al. 2015). Alternatively, thresholds could also draw on finding from

representative survey data; a threshold to separate the nutritionally empowered or disempowered could be

endogenous and computed based on whether it produces systematic differences in nutritional

outcomes/status between those that are empowered and those that are not.

We find that our threshold of 0.5 accurately reflects differences in nutritional status for BMI. For

haemoglobin levels, however, fitting a linear relationship between empowerment using a threshold of 0.5

and hemoglobin levels, did not yield statistically significant coefficients for the binary variable

representing nutritional empowerment. Increasing the threshold by 0.05 at a time, the classification of

those empowered and disempowered becomes significant statistically at 0.75 (Appendix 3). This suggests

that policies that aim to empower communities nutritionally need to recognize that marginal

improvements in measured nutritional empowerment might not necessarily be associated with significant

11

We do not test the threshold for individual indicators in this paper.

26

improvements in anemia. Indeed, calorie-protein adequacy may be enough for improving BMI whereas

improving anemia requires wider micronutrient adequacy and access to health care. If a goal is to improve

anemia status via nutritional empowerment, the aspirational threshold needs to be higher than the

threshold to achieve BMI improvements.

Fourth, we confirm that the list of indicators, I, used in the construction of WENI performs

comparably well with the full set of indicators (F) and that identified via factor analysis (S) on two

counts. First, I produces a rank order of individual nutritional empowerment scores similar to that

produced by F and S. We then compute the rank correlation of these the rank orderings. We use Kendall’s

tau-b to address the issue of ties in ranking. The rank correlations by scores suggest that rank orders are

maintained. Second, we compute rank correlation for each domain-dimension scores as a robustness

check and find that the ranking of individuals within each domain-dimension is also preserved going from

F to S to I (Table 8).

To gauge the sensitivity and consequences of paring down the set of indicators that constitute the

index, we estimate three separate equations using the entire set of indicators in F, S, and I. Table 9 shows

the R-squared, which is the proportion of total variation in the outcomes that are explained by the model,

for the sets F, S and I, for WEN women, spouses, and mothers-in-law. We run two versions of the model,

one without controls and the other with household and individual level controls, including house type,

roofing material, number of rooms in the house, religion, caste, age, and household wealth index of the

respondent. Not surprisingly, we find that the R-squared declines as we move from F to S to I.

Nonetheless, the total variation in BMI explained by the models using the I list of 40 indicators (with and

without controls) is 13 and 10 percent respectively for BMI and 37 and 42 percent for haemoglobin. In

terms of sub-groups, the figures for with controls and without are 14 and 19 percent for WENI women, to

24 and 30 percent for spouses, to 32 and 42 percent for mothers-in-law. For haemoglobin levels, we are

able to explain 8 and 19 percent for WENI women, 16 and 30 percent for spouses and 39 and 54 percent

for mothers-in-law, for with controls and without respectively. Despite the reduction in R-squared using

the smallest set of WENI indicators, the WEN models using I indicators still outperform existing models

27

relating empowerment measures and nutritional outcomes, indicating that the WEN grid variables are

predictive of nutritional status.12

8. Conclusion

Undernutrition rates among adults, and in particular women, in South Asia remain stubbornly

high. The WEN grid can provide insights into the domain-dimensions that contribute to gaps in

nutritional empowerment that impede improvements in nutritional outcomes. While recognizing that

empowerment is a process and the pathways to empowerment are ridden with obstacles, the WEN grid

can be a useful policy tool. When used as a scorecard, it allows examining the dimensions of

empowerment and domains of nutrition that are likely to bethe weak links and require policy attention. As

an index, it provides a comprehensive measure of nutritional empowerment that permits comparison

across socio-demographic groups, across administrative units or over time. As such, it allows

policymakers and practitioners to check progress relative to milestones on the path to nutritional

empowerment.

Efforts to design nutrition sensitive projects and policies might facilitate empowerment in one

area, for example agriculture, but, may not translate into empowerment of women in the realm of

nutrition. The WENI can potentially help gauge the nutrition sensitivity of such interventions. WENI is

also designed to be applicable for all rural households, including those not engaged in agriculture,

allowing for a more comprehensive understanding of the specific barriers and opportunities that rural

women and their families face. Further, many program that aim to inform and educate women on

nutritional issues presume women have power over allocation of resources and social practices are

conducive to favorable nutritional status, equally for all household member. WENI could offer a

systematic way to understand if these assumptions are correct.

12 As with other indices, the WENI too entails some loss of detail and trades off explanatory and

predictive power for parsimony. These tradeoffs limit the extent to which WENI can be simplified. With

adequate validation, there is scope to abridge WENI for a leaner version.

28

WENI is a summary measure but ,, like all indices, is also a statistical benchmark that represents

many complex and interdependent factors; it is therefore akin to an “information platform” (Alkire, et al.,

2013). Indeed, the WENI is not an end in itself but is best used as a heuristic device; it is important to

look beyond and behind the WENI to uncover and identify obstacles to nutritional empowerment that

deserve programmatic or policy attention. For this reason we do not advocate any comparisons across

communities and groups that relies exclusively on the WENI. Nor do we advocate using it to define

eligibility criteria for program participation.

The WENI presented here has been validated in two sites in India. We caution that the indicators

that ideally represent the component themes in WENI and their thresholds for empowerment may differ in

other contexts.. Nonetheless, the WENI is intentionally broad and aims to offer a complete articulation of

factors influencing nutritional empowerment.. The WEN grid allows for considerable latitude in choosing

specific indicators. There are many ways to capture key aspects of each domain dimension and retaining

this flexibility allows the WENI to be adapted to specific contexts. Future research will utilize data from

outside of South Asia to test whether the indicators identified for rural South Asia have salience in other

contexts.

29

References

Acock, A. C. 2013. Discovering structural equation modeling using Stata: Revised edition. Stata Press.

Alkire, S., Roche, J.M., Ballon, P., Foster, J., Santos, M.E. and Seth, S., 2015. Multidimensional poverty

measurement and analysis. Oxford University Press, USA.

Alkire, S., Meinzen-Dick, R., Peterman, A., Quisumbing, A., Seymour, G. and Vaz, A., 2013. The

women’s empowerment in agriculture index. World Development, 52, pp.71-91.

Alkire, S. and Foster, J. 2011.“Counting and multidimensional poverty mapping.” Journal of Public

Economics. 95(7-8): 476-487.

Black R.E., Victora C.G., Walker S.P., Bhutta Z.A., Christian P., de Onis M. et al. for the Maternal and

Child Undernutrition Study Group .2013. Maternal and child undernutrition and overweight in low‐

income and middle‐income countries. Lancet 382 (9890): 427–51. Maternal and Child Nutrition series.

Donald, A., Koolwal, G., Annan, J., Falb, K., & Goldstein, M. 2017. Measuring women's

agency. World Bank Policy Research Working Paper 8148.

FAO-IFAD-ILO, 2010.Gender dimensions of rural employment: differentiated pathways out of poverty.

Rome: United Nations

French Gates M. 2014. “Putting women and girls at the center of development. Science. 345(6202): 1273-

1275

Galiè, A., Teufel, N., Korir, L. et al. 2018. Women’s Empowerment in Livestock Index, Social

Indicators Research. https://doi.org/10.1007/s11205-018-1934-z

Headey, Derek, et al. 2015. "The other Asian enigma: explaining the rapid reduction of undernutrition in

Bangladesh." World Development 66: 749-761.

Kabeer, N. 1999. “Resources, agency, achievements: Reflections on the measurement of women’s

empowerment.” Development and Change. 30: 435-464.

Kadiyala, S. Harris, J., Headey, D., Yosef, S., and Gillespie, S. 2014. “Agriculture and nutrition in India:

mapping evidence to pathways.” Annals of the New York Academy of Sciences.1-14.

Kennedy, E. 1994. “Effects of sugarcane production in Southwestern Kenya on income and nutrition”.

von Braun J and Kennedy, E. (eds.) Agricultural commercialisation, economic development and nutrition.

Johns Hopkins University Press for the International Food Policy Research Institute: Baltimore, MD

Lentz, E., Narayanan, S., De, A. 2017. “Constraints to nutritional empowerment: Findings from multi-cite

research in South Asia.” Working Paper. Indira Gandhi Institute for Development Research.

Malapit, H.J.L., & Quisumbing, A.R. 2015. What dimensions of women’s empowerment in agriculture

matter for nutrition in Ghana? Food Policy, 52, 54-63.

30

Malapit, H.J.L., Kadiyala, S., Quisumbing, A.R., Cunningham, K. & Tyagi, P. 2015. Women’s

empowerment mitigates the negative effects of low production diversity on maternal and child nutrition in

Nepal. The Journal of Development Studies, 51(8), 1097-1123.

Meinzen-Dick, R. 2013. “How I learned to stop worrying and love the Index: Women’s Empowerment in

Agriculture.” International Food Policy Research Institute Reflections. April. 1-4.

Narayanan, S., Fontana, M., Lentz, E., & Kulkarni, B. 2017a. Rural women’s empowerment in nutrition:

A proposal for diagnostics linking food, health and institution, Background Paper for UN Women,

Commission on the Status of Women (CSW) 62.

Osborne, J.W., Costello, A.B. and Kellow, J.T., 2008. Best practices in exploratory factor analysis. Best

practices in quantitative methods, pp.86-99.

Pratley, P. 2016. “Associations between quantitative measures of women’s empowerment and access to

care and health status of mothers and their children: A systematic review of evidence from the developing

world.” Social Science and Medicine. 169: 119-131.

Robeyns, I. 2003. “Sen’s capability approach and gender inequality: selecting relevant capabilities.”

Feminist Economics. 9(2-3): 61-92.

Sraboni, E., Quisumbing, A., Ahmed, A. 2013. “The Women’s Empowerment in Agriculture Index:

Results from the 2011-2012 Bangladesh Integrated Household Survey.” April. International Food Policy

Research Institute Report.http://www.ifpri.org/sites/default/files/publications/weaitr.pdf

Tabachnick, B.G. and Fidell, L.S. 2007. Using Multivariate Statistics (5th ed.). New York: Allyn and

Bacon.

UN Women. 2018. Turning promises into action: Gender equality in the 2030 Agenda for Sustainable

Development. New York: United Nations

UN Women. 2014. World Survey on the Role of Women in Development; Gender Equality and

Sustainable Development. New York: United Nations.

UNICEF (1990).Strategy for improved nutrition of children and women in developing countries.

UNICEF Policy Review 1990‐1 (E/ICEF/1990/ L.6). New York: UNICEF

31

Table 1: Themes in the WEN Grid

Food knowledge (n= 2) Food resource (n=9) Food agency (n=6)

Knowledge of calcium

Knowledge of iodine

(1) Participation in income generating activities

Does paid work as employee

Participation in non-agricultural household enterprise Participation in agriculture and allied activities

(2) Access to support and to assets

Aid for crop and livestock production. Financial support for business

Access to food aid schemes and employment schemes

Asset owned by individual (3) Eating norms

No imposed dietary restrictions

Eating order (all eat together)

(1) Say in productive activities

Say in cultivation decisions

Say in kitchen garden production Say in major/minor decisions related to household enterprise

(2) Control /say in income and expenditure

Say in rent or sale of own assets Say in expenditure of own contribution to income

Say in purchasing food items

Health knowledge (n=3) Health resource (n=5) Health agency (n=3)

Knowledge of anemia Knowledge of ORS

Knowledge of malaria

(1) Work/energy expenditure and working conditions Does no activity, paid or unpaid, which is back breaking or heavy

Does no activity paid or unpaid, with risk of injury or risk of major health problem

(2) Support in work Has help with domestic and care work; does not undertake care work

(3) Support when ill and health seeking

Not ill in past 30 days (chronic or otherwise) or sought treatment when ill Has help for housework when ill

(4) Access to improved water, sanitation and smoke free kitchen

Can go alone to health centre when required Does not need to take permission for going to health centre

Takes decisions on own health

Institution (n=7)

Membership: Member or non-membership of groups out of their own volition

Access to information about government schemes

Intimate partner violence: Does not experience physical abuse or gets family support when experiencing physical abuse.

Freedom of movement: Has visited bank/post office unaccompanied/can visit family alone No risk of sexual harassment in paid/other work

Veil use: Does not practice burkha

Public/civic engagement: Participated in rally/protests/petition/Comfort speaking in public

Fertility knowledge (n=2) Fertility resource (n=5) Fertility agency (n=4)

Knowledge of menstrual cycle and

pregnancy Knowledge of need of different diets

during pregnancy

Ability to secure an adequate, appropriate and diverse diet during pregnancy

Did not undertake heavy physical activity after 8th month of pregnancy Did not do heavy physical activity for at least a month after delivery

Received support in housework during pregnancy

At least one antenatal and postnatal care facility of good quality accessible and affordable

Has a say in use/not use of contraceptive

Has a say in the choice of facility for delivery Has a say in total no of children

Has a say in spacing of children

32

Table 2: Summary Statistics of Nutritional Empowerment Indicators

Domain

Dimension

(Number of

indicators)

Description of the indicator Odisha Bihar Total

WENI

Woman Spouse

Mother-

in-law

WENI

Woman Spouse

Mother-

in-law

FOOD

Food

Knowledge

(2)

1 Knowledge of foods rich in calcium

6.2 8.9 1.4 14.8 13.2 4.4 9.8

2 Knowledge that iodized salt is a key source of iodine

18.5 40.0 7.0 29.0 33.8 6.7 25.1

Food

Resources

(9)

1 Does market work as employee, for wages or salary, in cash or kind 32.6 76.3 22.5 36.0 63.6 18.9 42.5

2 Has at least three distinct types of food resources available via

participation in the following: ruminants, poultry, fishing, aquaculture,

kitchen gardens 48.5 37.0 39.4 56.9 58.3 55.6 51.0 3 At least 50% of village reports its acceptable for women to work in village

96.5 98.5 97.2 100 100 100 98.8 4 Owns land in his or her own individual capacity

20.7 21.5 21.1 3.7 16.6 11.1 14.1 5 Household members eat together

43.2 56.3 39.4 9.8 4.0 7.8 25.1 6 Has diverse sources of income (at least 3 of 5, from the following: paid

work, own account family enterprise, agriculture, transfers, remittances) 7.1 10.4 5.6 18.2 30.5 20.0 15.7 7 Own account worker/contributing family worker in agriculture or non-

agriculture family enterprise 67.8 61.5 49.3 63.0 84.1 56.7 65.6 8 Is eligible to access state food subsidies and workfare program

57.3 61.5 64.8 22.6 27.8 25.6 40.3 -Proportion of households who have a ration card

84.6 83 84.5 76.4 76.8 81.1 80.3 -Proportion of households who have a job card

65.6 72.6 76.1 27 32.5 30 47.1 9 Financial support for either business or agriculture

24.2 21.5 11.3 50.5 22.5 20.0 30.3

Food

Agency

(7)

1 Decision to undertake or not undertake paid work own 37.9 86.7 76.1 45.5 72.2 85.6 59.5

2 Has no dietary restrictions not of his/her own volition and can give these

up if he/she wishes 22.5 51.1 28.2 14.8 26.5 18.9 24.8 -Proportion who have no dietary restrictions

10.1 30.37 16.9 4.04 11.3 5.6 11.3 3 The earnings from an asset owned by the respondent has not been used

without his/her consent 21.6 57.8 47.9 15.2 21.2 23.3 26.7 4 At least some say in kitchen garden decisions and/or cultivation

35.2 46.7 32.4 36.7 52.3 47.8 40.9 5 At least some say in major or minor own account family enterprise

decisions 10.6 20.0 5.6 6.7 30.5 1.1 12.6 6 Has independent source of cash (earned, transfers or remittances) and has

some control over how to use it 74.9 90.4 59.2 92.6 99.3 91.1 86.6 7 Has a say in how his/her contribution to family income is to be spent

38.3 90.4 28.2 36.0 86.1 26.7 50.5

33

HEALTH

Health

Knowledge

(3)

1 Can recognize if someone has anaemia 19.4 14.1 8.5 35.0 39.1 30.0 26.7

2 Aware of what ORS is for and when to administer it 68.3 68.1 47.9 29.6 43.7 23.3 47.0

3 Aware of how malaria is transmitted 63.4 68.9 42.3 46.5 62.9 31.1 54.4

Health

Resources

(9)

1 Has no chronic illness in the past year or illness in the past month; or has

either but sought health care 85.5 94.8 88.7 93.6 96.7 95.6 92.2

- Proportion who did not have any illness in the past one month 41.8 28.9 39.4 60.6 42.4 57.8 47.2

-Proportion who were unwell in the past one month 58.2 71.1 60.6 39.4 57.6 42.2 52.8

-Proportion with chronic illness in the past one year 15 14.1 49.3 19.9 19.2 45.6 22.4

2 At most one type of paid work undertaken 89 80 94.4 91.9 74.2 96.7 87.4 - Proportion who does not undertake paid work

67.4 23.7 77.5 63.9 36.4 81.1 57.5 - Proportion who does one type of paid work

22.0 56.3 16.9 28 37.8 15.6 30 3 Very satisfied with the time available to visit friends and family or living

with them 12.8 42.2 42.3 16.2 25.8 32.2 23.9 - Proportion who lives with their family

4.8 97 8.5 2.4 79.5 6.7 28.9 4 Does no productive work (whether paid or unpaid) activity that is back

breaking or heavy 66.5 23.7 74.6 62.6 35.8 81.1 56.5 - Proportion of those who do paid work

32.6 76.3 22.5 36.1 63.6 18.9 42.5 Proportion of the above who says the work is heavy or back

breaking 87.8 97.1 93.8 92.5 88.5 100 92.3 - Proportion of those who work in family owned non-agricultural

enterprises 14.1 20 7 7.4 31.3 1.1 13.8 Proportion of the above who says the work is heavy or back

breaking 68.8 33. 3 60 59 34 0 47 5 At least somewhat satisfied with help in domestic & care work

76.7 100 98.6 89.9 100 97.8 91.1 6

Gets assistance in domestic and care work’ when ill 80.6 94.8 84.5 77.1 90.7 100 85.2 7 Household has access to improved water, improved toilets & a well

ventilated cooking area 14.5 17.8 22.5 39.1 39.7 47.8 30.1 -Household has access to improved water

75.3 74.8 80.3 98 98.7 97.8 88.3 -Household has access to improved sanitation

33.5 39.3 42.3 42.1 41.7 50 40.4 -Household has access to ventilated cooking area

50.2 51.9 54.9 92.9 95.4 94.4 75 8 Productive work (paid or hh enterprise) <= 8 hours

81.1 52.6 90.1 86.5 48.3 95.6 75.7 9 Faces no risk of injury or major health problem in any activity

50.7 14.8 59.2 44.4 17.2 58.9 40.0

Health

Agency

1 Can visit the health facility alone if needed 35.7 95.6 23.9 54.9 98.0 30.0 58.2

2 No expectation to seek permission from the family before visiting health 38.3 69.6 42.3 26.6 84.8 43.3 47.1

34

(3) facility

3 Can make decision on own health 17.6 90.4 8.5 18.9 47.0 10.0 31.3

INSTITUTIONS

Institutions

(7)

1 Can become a member of any group out of his/her own accord 13.7 30.4 25.4 20.5 17.9 37.8 21.8

2 Never uses a veil 22.0 100 14.1 2.4 100 7.8 37.1

3 Has a mobile phone and receives information on government schemes 15.4 12.6 5.6 24.6 29.1 3.3 18.1

4 Does not experience domestic violence or experiences domestic violence

but has support 78.0 93.3 81.7 91.2 95.4 95.6 88.8 - Proportion who does not experience domestic violence

62.6 92.6 77.5 66 91.4 93.3 76 5 Has freedom to visit bank, post office, or family alone

54.6 99.3 56.3 69.0 100 75.6 74.4 6 Does not experience sexual harassment in any activity, paid or unpaid

99.6 100 98.6 96.3 94.0 98.9 97.6 7 Civic Engagement: Has participated in at least one activity in the last five

years (petition, protest, public meetings, representation to government

officials, voted in elections) 91.2 98.5 97.2 86.5 97.4 98.9 92.9

FERTILITY DOMAIN

Fertility

knowledge

(2)

1 Has knowledge of menstrual cycle & likelihood of pregnancy

4.9

12.3

9.1

2 Is aware that pregnant women have different dietary needs

76.4 91.1 84.7

Fertility

Resources

(5)

1 Cash or in-kind assistance from Govt. during pregnancy

93.3

68.3

79.2

2 Did not undertake heavy physical activity after 8th month of pregnancy

11.1

9.6

10.2

4 Received at least a fair bit of support from family during pregnancy

43.6

44.1

23.7

5 Has access to natal care facility of good quality & very affordable

93.8

71.4

43.8

6 Had adequate and diverse diet during last pregnancy

19.8

8.4

7.2

Fertility

Agency

(3)

1 Place of delivery decided based on respondent's preference

89.3 75.8 81.7

2 Has at least some say in total number of children

49.8

64.8

58.3

3 Has at least some say in spacing between children

67.0 74.4 38.4

Note: The figures reported by WENI, Spouse and MIL are not comparable even when the indicator is measured at the household level, since not all spouses and MILs were

interviewed.

35

Table 3: Health and Nutritional Outcomes in the Sample

Odisha Bihar All

WENI Spouses MIL WENI Spouses MIL Total

Nutritional status

Mean body mass index (BMI) 21 22 21 20 21 21 21

Underweight (%) 31 16 22 29 14 30

25

Mean haemoglobin levels (mcg) 11 13 10 11 13 11 11

Mild anemia (%) 26 36 10 23 32 20 26

Moderate anemia (%) 54 8 61 43 5 42 36

Severe anemia (%) 3 1 13 4 1 1

3

Mild anemia or worse (%) 82 44 83 71 37 63

65

Moderate anemia or worse (%) 56 9 73 48 5 43

39

Health outcomes

Unwell in the past month (%) 42 29 39 61 42 58 47

Suffering from chronic illness in

past 12 months (%) 15 15 49 20 19 46

22

36

Table 4: WENI, Headcount Ratio and Other Measures

POOLED Without Fertility Domain Only WENI Women (with fertility)

Odisha Bihar All Odisha Bihar All

WENI Spouses MIL WENI Spouses MIL

Nutritional

Empowerment Index 0.417 0.574 0.380 0.421 0.552 0.399 0.457 0.450 0.454 0.452

Head count ratio of

nutritionally

disempowered (%) 79 29 89 75 40 81 66 70 69 69

Intensity of

disempowerment 0.603 0.432 0.649 0.615 0.479 0.607 0.567 0.304 0.313 0.309

Gap in Empowerment

score 0.125 0.046 0.140 0.133 0.067 0.137 0.110 0.103 0.102 0.102

Adjusted Headcount Ratio

(%) 48 13 58 46 19 49 39 21 22 21

Adjusted Gap 0.059 0.006 0.081 0.062 0.013 0.067 0.048 0.022 0.022 0.022

Note: The nutritional empowerment index is the proportion of total domain-dimensions in which an individual is deemed empowered.

Head Count Ratio (HCR) is the percentage of those classified as nutritionally empowered – i.e., if their nutritional empowerment index is greater than or equal to

0.5. This would mean that the individual is deemed empowered in at least 50% of the seven domain-dimensions that constitute WENI and 10 domain dimensions

for WENI women.

37

Table 5: Domain and Dimension-wise Disempowerment Headcount Ratio

Odisha Bihar All

WENI

Women Spouses

Mothers

in law

WENI

Women Spouses

Mothers

in law

Domain (in

%) Food 78 48 80 79 53 88 71

Health 37 17 31 39 33 29 33

Institution 40 0 39 33 0 19 24

Fertility 28

34

63

Dimensions

(in %) Knowledge 77 62 90 74 62 84 73

Resource 36 33 13 31 42 12 31

Agency 64 7 70 60 3 57 45

Knowledge (incl Fertility) 75

72

86

Resource (incl Fertility) 44

55

73

Agency (incl Fertility) 58 53 76

Note: All refers to a weighted average of the headcount ratio for both the states of Bihar and Odisha

38

Table 6: Empowerment Status and Outcomes: Validity Tests

Disempowered Empowered

Difference T / Z

statistic

p-value

Mean

/Proportion N

Mean

/Proportion N

Test of Equality of Means

BMI 20.35 608 22.08 320 -1.74 -7.04 0

Hb levels 11.06 578 12.23 296 -1.17 -8.63 0

Test of Equality of Proportions

Normal BMI 0.7 608 0.83 320 -0.14 -4.6 0

No anemia 0.25 578 0.42 296 -0.17 -5.25 0

39

Table 7: The Relationship Between Nutritional Empowerment and Status (Full Sample & WENI Women only)

Variables BMI Normal BMI(=1) Hb levels Anemia (=1)

Nutritionally empowered

(=1)

1.617***

0.330***

0.204

0.102

(5.73)

(2.99)

(1.56)

(0.96)

Nutritional

empowerment index

7.113***

1.569***

1.118**

0.660

(6.67) (3.67) (2.21) (1.59)

Spouse (=1) 0.409 0.106 0.446*** 0.371*** 2.271*** 2.203*** 0.914*** 0.868***

(1.37) (0.34) (3.60) (2.87) (15.13) (14.04) (7.96) (7.27)

Mother-in-law (=1) 0.197 0.455 0.284 0.352 -0.274 -0.224 0.005 0.035

(0.29) (0.67) (1.17) (1.44) (-0.92) (-0.75) (0.02) (0.14)

Age (completed years) 0.017 0.012 -0.005 -0.006 -0.002 -0.003 -0.004 -0.005

(0.92) (0.61) (-0.68) (-0.90) (-0.30) (-0.44) (-0.59) (-0.69)

Constant 19.588*** 17.149*** 0.560*** 0.025 10.843*** 10.455*** -0.708*** -0.942***

(33.79) (25.25) (2.60) (0.10) (45.56) (35.11) (-3.36) (-3.63)

N 928 928 928 928 874 874 874 874

R-squared 0.058 0.069

0.324 0.326

Adjusted R-squared 0.054 0.065 0.321 0.323

WENI Women Only

Nutritionally empowered

(=1)

1.414***

0.350***

-0.022

0.024

(3.79)

(2.65)

(-0.14)

(0.17)

Nutritional

empowerment index

4.478***

0.808

0.274

-0.055

(2.81) (1.40) (0.35) (-0.09)

Age (completed years) 0.033 0.035 0.003 0.004 -0.001 -0.002 -0.003 -0.003

(1.01) (1.12) (0.24) (0.36) (-0.11) (-0.17) (-0.27) (-0.25)

Constant 19.064*** 17.398*** 0.306 0.003 10.862*** 10.751*** -0.725** -0.699*

(19.86) (15.88) (0.87) (0.01) (33.64) (24.66) (-2.28) (-1.77)

N 494 494 494 494 474 474 474 474

R-squared 0.0403 0.0233 0.0003 0.0001

Adjusted R-squared 0.0122 0.0039 0.0002 0.0001

40

Table 8: Comparison of the Full Set of WEN Grid Variables with Indicators Included in the Index

WENI Women Pooled

From F

to S

From S

to I

From F

to S

From S

to I

3.1.a Food Knowledge 1.00 1.00 1.00 1.00

3.1.b Food Resource 0.65 0.63 0.60 0.59

3.1.c Food Agency 0.51 0.63 0.49 0.63

3.1.d Health Knowledge 1.00 1.00 1.00 1.00

3.1.e Health Resource 0.48 0.35 0.41 0.34

3.1.f Health Agency 0.43 0.80 0.58 0.86

3.1.g

Ferility Knowledge (only

WEN) 1.00 1.00

3.1.h

Fertility Resource (only

WEN) 0.54 0.38

3.1.i

Fertility Agency (only

WEN) 0.76 0.83

3.1.j Institutions 0.57 0.52 0.61 0.59

3.2 Food Domain 0.58 0.70 0.55 0.67

3.3 Health Domain 0.55 0.44 0.52 0.47

3.4

Fertility Domain (only

WEN) 0.56 0.61

3.5 Knowledge Dimension 1.00 1.00 1.00 1.00

3.6 Resource Dimension 0.48 0.39 0.39 0.25

3.7 Agency Dimension 0.54 0.74 0.49 0.74

3.8 Overall WEN grid 0.58 0.60 0.55 0.62 Note: The number of indicators for WEN women is greater than the number of pooled indicators because the pooled

indicators exclude fertility questions.

41

Table 9: R-squared from regressing BMI and Haemoglobin Levels on F, S and I

F S I

Controls No controls Controls No controls Controls No controls

Least squares for BMI and haemoglobin levels

BMI

All 0.20 0.26 0.12 0.18 0.10 0.13

WENI Women 0.39 0.36 0.25 0.21 0.19 0.14

Spouses 0.68 0.61 0.35 0.32 0.30 0.24

MIL - 0.91 0.47 0.32 0.42 0.32

Haemoglobin

All 0.52 0.50 0.43 0.38 0.42 0.37

WENI Women 0.41 0.34 0.24 0.15 0.19 0.08

Spouses 0.72 0.66 0.41 0.24 0.30 0.16

MIL - 0.99 0.57 0.42 0.54 0.39

Linear probability models for binary outcomes

Normal BMI

All 0.20 0.18 0.12 0.10 0.10 0.07

WENI Women 0.29 0.27 0.14 0.11 0.10 0.07

Spouses 0.59 0.50 0.25 0.19 0.23 0.13

MIL - 0.89 0.44 0.32 0.39 0.28

No anemia

All 0.36 0.32 0.27 0.21 0.24 0.18

WENI Women 0.40 0.34 0.22 0.14 0.16 0.07

Spouses 0.69 0.65 0.37 0.27 0.29 0.21

MIL - 0.97 0.54 0.43 0.52 0.39

* Note the total number of observations for these regressions under set F is only marginally higher than

the parameters estimated and the R-squared should therefore be interpreted with caution. For this reason,

results for the Mother in law are not reported for F. In addition to WEN grid elements, “Controls”

regressions include other household and individual level controls. These are house type, roofing material,

number of rooms in the house, physiological status, religion, caste, age and household wealth index of the

respondent.

42

Online Supplementary Materials

Appendix 1: Construction of indicators

The indicators within each domain-dimension in Table 1 represent a substantive theme or aspect

of empowerment that might matter for nutrition, but do not prescribe specific questions, which would

depend on the context and the sub population studied. On the one hand, the WENI allows flexibility in

choice of indicators depending on the context and researcher, programmatic or participant priorities. For

example, in the context of freedom of movement, where most women have such freedoms – for example,

to visit family, the market or a bank, or the ability to drive a vehicle might represent greater

empowerment. Another illustration of contextual specificity pertains to decisions around cultivation and

kitchen gardens. These are more relevant for agrarian communities and own land or homesteads. In

contrast, where markets dominate as the most important source of food, how much say the individual has

in procuring, processing and cooking might matter more. On the other hand, one might simply want to

use a uniform set of indicators across contexts to both enable comparisons.

In any index, it is challenging to capture indicators that have complex or ambivalent values. Yet,

to construct the index, it is necessary to assign cut-offs indicating whether responses are empowering or

disempowering for each specific indicator. For example, women’s access to paid work can increase both

their access to resources and enhance agency within the household. At the same time, if such work is

hazardous or poorly paid, it might be harmful as well. Similarly, many individuals in resource constrained

settings take up multiple jobs. While this strategy helps smooth earnings over time, it also might indicate

poor quality, fragile jobs that might reduce leisure or increase stress. We attempt to capture both aspects

in WENI. Access to paid work and participating in plural activities is an indicator for Food Resource, but

having at most one type of paid work is deemed an empowering Health Resource.

The coding of responses to some indicators as empowering or disempowering are more open to

debate. For example, our indicator capturing Intimate Partner Violence (IPV) assign to women a score of

1 if they do not face domestic violence at all or if they do they have support to cope with it. One could

argue that the former is the better measure of empowerment and it is incorrect to admit the possibility that

women who experience IPV are deemed empowered just because they have support. In the WENI

presented here, we chose both forms of the indicator as candidates, but only one of them emerged relevant

in the data reduction exercise.

We also constructed indicators to be free from gender specificities in order to ensure direct

comparability and scale equivalence across genders. On the use of veil for example, we ask both men and

women this question, even though this question is not applicable to men. Similarly, in most contexts

43

women are the majority of IPV victims. We, however, ask men for sources of domestic conflict, drivers

of these conflicts and incidence of violence against them. We found a few cases where men experienced

violence within the household.

Our formative research in Bangladesh and India suggests that state support could significantly

influence the ability of women to access resources. The index does a modest job of capturing the full

extent of support the state. The WENI captures explicitly government schemes to provide work and in-

kind rations to the poor, assistance to pregnant women. Household access to water, toilets and sanitation,

do not reflect the specific role of the state but could reflect public investments in these areas. Similarly,

health and nutritional knowledge could be a consequence of government interventions. The extent to

which state role in forwarding nutritional empowerment is captured depends on the research question.

Many indicators in WENI are captured at the household level (access to toilets, water, etc.) that

can be better captured at the individual level. Other indicators are subjective assessments and as such

inherit the problems associated with most subjective measures. These choices might not be ideal, but we

chose these to reduce the burden of interviews on the respondents.

Thus overall, while WENI’s candidate themes and indicators are meant to capture

comprehensively our conceptualization of nutritional empowerment, we do not define a list of indicators

that are appropriate for all contexts. Nor is there a perfect approach to capture these that also allows us to

minimize the burden of surveys for the respondents. An index based on this is sure to be sensitive to the

inclusion and exclusion of variables, as also to forms in which the question is administered and the cut-

offs used. We discuss findings regarding indicator selection in the section on robustness checks.

44

Appendix 2: The WENI Survey

The SPANDAN survey was funded by the Bill and Melinda Gates Foundation and was conducted

collaboratively by the Indira Gandhi Institute for Development Research, Mumbai India and the National

Institute of Nutrition, Hyderabad, India. For more information on SPANDAN’s coverage, questionnaires,

and sampling procedures, see: http://www.spandan-india.org/Survey/Survey.aspx. SPANDAN surveyed

members of households that included a mother of at least one child under the age of 5. We drew our

sample of WENI women from the same set of SPANDAN mothers. Thus, the sampled WENI women had

at least one child age seven or under, with the exception of a few WENI women whose children had died.

We selected each of the districts covered in the SPANDAN survey. These districts were

originally randomly selected from each of nine agro-climatic zones as identified by Government of

India’s National Agricultural Resource Program. Within SPANDAN’s selected districts, we drew a

subsample of one SPANADAN surveyed block in each district. Within each of the blocks, we randomly

selected a SPANDAN surveyed village that had been interviewed during SPANDAN’s Round 1 and we

randomly selected a SPANDNA surveyed village that had been interviewed during SPANDAN’s Round

2. In total, the WENI sample was drawn from 9 districts (5 in Odisha, 4 in Bihar), 9 blocks, and 18

villages. Within the sample villages, our goal was to interview all previously surveyed mothers in

SPANDAN households. Due to attrition, we were unable to interview 21.5% of women. Reasons for

attrition include non-availability of the WENI women due to work, hospital visits, attending weddings,

and visiting to parents. In a few cases, either the woman declined to be interviewed or her family did not

give her permission to be interviewed. In a few other cases the woman was deceased or mentally or

physically unwell.

To compensate for the attrition and due to the small number of SPANDAN households within

each village, we interviewed women who were surveyed by SPANDAN in another SPANDAN village in

the same block. In all, we covered 26 villages, as opposed to the original 18, choosing 8 additional

villages to expand the households covered. To capture the relative disparities within households, we

sampled SPANDAN households where WENI Women lived with her male spouse and or her mother-in-

law. If the woman interviewed in the SPANDAN survey was not available, the entire household was

skipped. If the husband or mother-in-law was selected to be interviewed but was unavailable, the WENI

Woman was interviewed and the mother-in-law or husband was interviewed in the next selected

household. In nine SPANDAN households (eight in Bihar and one in Odisha), there were had two

mothers with children under age five. The WENI Survey interviewed whoever was listed first on the

SPANDAN household roster, and interviewed the second mother only if the first was unavailable. If this

45

household had split after the SPANDAN survey such that the two mothers were now part of separate

households, then both were interviewed.

Dried blood spot samples

The dried blood spot samples were processed at the laboratory facilities at the National Institute

of Nutrition (NIN, Hyderabad, India). A qualified practitioner provided respondents information on their

height, weight and dried blood spot test result. For respondents with low hemoglobin levels, the

practitioner also provided advice on the treatment of anemia, including advising those with severe

anaemia to visit the nearest public health centre as soon as possible. We also provided the answers to the

nutritional knowledge questions asked in the survey.

Appendix Table 1: Distribution of the Sample Agro climatic

Zone

District and State Block (Villages) Number of respondents ( Number for

whom Haemoglobin was measured)

WENI

women

Spouses Mothers in

law

1 Northern

Plateau

Mayurbhanj, Odisha Baripada (2), Khunta(1) 67 (64) 41 (36) 21 (18)

2 Coastal Zone Kendrapara, Odisha Patkura (2), Mahakalapada

(1)

57 (55) 30 (30) 28 (25)

3 Central Table

Land

Anugul, Odisha Athamalik (2), Talcher

Sadar (1)

31 (29) 24 (22) 6 (6)

4 Eastern Ghats Rayagada, Odisha Muniguda (1), Ambdala

(1), Padmapur (1)

37 (34) 17 (17) 6 (6)

5 Western

Undulating

Nuapada, Odisha Boden (2), Jonk (1) 35 (35) 24 (23) 10 (10)

6 North-East

Alluvial Plains

Purnia, Bihar Dagarua (1), Banmakhi (2) 141

(138)

77 (72) 40 (34)

7 North-West

Alluvial Plains

Sitamarhi, Bihar Charaut (2), Pupri (1) 51 (46) 23 (22) 16 (16)

8 South Bihar

Alluvial Plains

(a)

Jamui, Bihar Jamui (2), Lakshmipur (1) 82 (77) 43 (36) 25 (21)

9 South Bihar

Alluvial Plains

(b)

Rohtas, Bihar Tilothu (1), Akhori Gola

(1)

21 (19) 11 (10) 9 (8)

46

Appendix 3: Sensitivity to thresholds and outcome measures

This table presents the coefficient of WENI under different cutoffs. Some windows have sparse

observations or have coefficients of 0 and are not reported. Full results are available with the authors.

Threshold BMI Normal weight (=1) Hb level No anemia (=1)

0.5 1.546*** 0.095*** 0.211* 0.109

0.55 1.617*** 0.080** 0.167 0.095

0.6 1.469*** 0.041 0.071 0.106

0.65 1.283*** 0.061 0.148 0.214

0.65 2.154*** 0.045 0.318 0.263

0.7 0.987 0.172*** 0.526* -

0.75 2.304 0.156*** 0.466** 0.755**

0.8 6.182*** 0.146*** 0.338*** -

47